BACKGROUND & AIMS: Liver failure has high mortality. There are accurate but controversial models to determine mortality of patients with acute-on-chronic hepatitis B liver failure (ACHBLF). We developed a logistic regression model (LRM) and tested its ability to predict 3-month mortality of patients with ACHBLF.
METHODS: The LRM was constructed using data from an internal cohort of 242 consecutive patients with ACHBLF and tested on an external cohort of 210 patients with the same conditions. The receiver operating characteristic curves (ROCs) were calculated for the LRM, model of end-stage liver disease (MELD), Child-Pugh score (CPS), and a previously reported LRM that has not yet validated in patients with ACHBLF. Predictions of 3-month mortality obtained with 4 models from the same datasets were compared using areas under ROC curves.
RESULTS: The LRM identified 5 independent factors associated with survival of patients with ACHBLF: hepatic encephalopathy (odds ratio [OR] = 2.165, 95% confidence interval [CI], 1.015-4.616), hepatorenal syndrome (OR = 9.767, 95% CI, 4.273-22.328), liver cirrhosis (OR = 2.339, 95% CI, 1.110-4.930), hepatitis B e antigen (OR = 2.874, 95% CI, 1.376-6.003), and prothrombin activity/age (OR = 0.12, 95% CI, 0.037-0.395). Data from the internal and external cohort indicated that the LRM had significantly greater prognostic accuracy than the MELD, CPS, or previous LRM.
CONCLUSIONS: We developed a logistic regression model that predicted 3-month mortality of patients with ACHBLF with greater accuracy than the MELD, CPS, or the previous LRM.